1. Field of the Invention
This invention relates to database management systems and more particularly to a system and method providing support for long-term storage and retrieval of objects created by application programs written in object-oriented programming languages.
2. Description of Related Art
Many new computer software applications, such as Computer-Aided Design and Manufacturing, Computer-Aided Software Engineering, multimedia and hypermedia information systems, and Artificial Intelligence Expert systems, have data models that are much more complex than previous systems, both in content and inter-object relationships. Object-oriented languages provide the application developer the mechanism to create and manipulate the data models inherent in these applications. Database systems provide long term storage of the data created by these applications. However, existing languages and databases are insufficient to develop these applications because existing object-oriented languages do not provide direct support for long-term storage and sharing of objects, existing commercial database systems (hierarchical, network, and relational) do not support the necessary complex object-oriented data models, and existing database systems require an application developer to use different languages and modeling paradigms when building applications.
There have been various research and commercial efforts aimed at developing OODBs. These OODBs vary in type of data model employed, application program interaction, object access method, method of persistent object store, etc. Examples of these current OODBs, and their weaknesses, will now be considered.
Iris (Hewlett Packard) and GemStone (Servio Corporation) are representative OODBs employing new proprietary object-oriented data models, while Vbase (Ontologic), Orion (Microelectronics and Computer Technology Corporation), and DOME (Dome Software Corporation) are examples of OODBs incorporating proprietary extensions to existing programming language data models. In these types of OODBs, application developers are required to learn a new proprietary data model in order to effectively use the OODB. Since their data model is new, using it often results in a loss of productivity as application developers learn the new language. In Orion, instances of user-defined classes cannot be stored in the database unless they have been derived from Orion-defined classes. In addition, GemStone and Orion do not allow for an instance of their classes to be transient; that is, every object created in a GemStone- or Orion-based application will be stored in the database unless it is specifically deleted. Another problem with developing a new data model is that it requires application developers to rely on a single source of application development tools, such as language compilers, object libraries, and program debuggers, which limits widespread acceptance of these OODBs.
POSTGRES (University of Berkeley) is an example of an OODB employing another type of data model, that of a proprietary extension to an existing relational database. POSTGRES is a combination of an extended relational and object-oriented database. Objects are created using relational table descriptions, while functions to manipulate the objects are created using the POSTQUEL query language as well as conventional languages (C and LISP). In addition, if the application developer wishes to add indices over user-defined types, they must write and register with POSTGRES functions to perform the various comparison operations between two objects of the same user defined type. Since the latter mode of creating functions requires the application developers to map between the POSTGRES and C/LISP data models, which can be error prone and distracting from the task of developing the application system, this strategy does nothing to alleviate the burdensome requirement to use different languages and modeling paradigms when building applications. This problem of mapping between the object-oriented and relational data models was discussed in-depth in the Intermedia OOPSLA '87 conference paper.
Ontos(Ontologic) and Object Store(Object Design) are representative of OODBs employing the last type of data model, namely the use of an existing programming language data model (e.g., using the C++ programming language data model for writing software programs and interacting with the database). Both systems, however, require the use of a proprietary language compiler to add additional code (Ontos) or translate new and non-standard C++ language constructs (Object Store). As with the first two types of OODBs, this approach requires application developers to rely on a single source of application development tools, which also limits widespread acceptance of these OODBs.
In addition to problems inherent with the type of data models selected, difficulties occur when an application program interacts with an OODB. In the Iris OODB, application developers define object types and develop functions to manipulate the objects using the proprietary Iris language. Iris provides an interactive interface where requests can be made to retrieve or manipulate Iris objects. The requests are evaluated by performing relational queries (since the objects are stored in relational tables) and the result is returned as an Iris expression, not as object values or references. Iris provides an embedded object SQL interface, a C language interface (which is not object-oriented), and allows the application developer to register foreign functions written in existing (possibly non-object-oriented) programming languages. These approaches require the application developer to map the Iris objects into data structures accessible by the programming language, reintroducing the problems discussed above.
Similarly, the developers of the GemStone OODB also defined a new language, OPAL, which the application developer uses to define object types and functions to manipulate the objects. GemStone provides an interactive development environment for developing OPAL objects and functions. GemStone also provides a mechanism for existing programming languages (C and Smalltalk) to interact with GemStone. However, unless the applications developer uses only the OPAL language, two data models and languages must be used to interact with the database, mapping the OPAL objects into structures accessible by the programming language, and thereby resulting in the problems associated discussed previously.
The Vbase OODB requires two separate languages, TDL to define object types, and COP (an extension to the C programming language) to develop application programs. Although application developers do not need to map objects between the data model and the programming language, they must still use two languages during the development of their programs, with the attendant problems considered above. A further restriction of this system includes the failure to provide access to the database from other programming languages.
Although the Orion OODB developers used an existing programming language, Common Lisp, for their data model, they developed several proprietary extensions to the language. As with Vbase, there is no need to map between the data model and programming language with the Orion OODB. However, this approach requires the use of a proprietary language translator.
The developers of POSTGRES, on the other hand, expect most application developers to write programs that interact with file database primarily using the POSTGRES query language, POSTQUEL. Navigation between objects is possible; however, a query must be issued to perform the navigation instead of accessing the referenced object directly. Application developers can define and implement their own functions including programming language statements, POSTQUEL query statements, and/or calls to POSTGRES' internal functions. Thus, application developers may have to deal with two or more data models to build their application systems. Such requirement fails to alleviate the problems considered above.
The Ontos approach provides an interface from the C++ language to the database. However, the amount of interaction between the program and Ontos is much higher than is reasonable or necessary due to the requirement of specialized functions that must be provided by the application developer (e.g., object construction, translation, storage/retrieval, etc.). This burdens the application developer with more work that could have been performed by the database system. Object Store also provides an interface from the C++ language to the database. However, the interface is accomplished by redefining the semantics of or adding new C++ language constructs, thereby requiring the use of Object Design's proprietary C++ language translator, which limits widespread acceptance of their system.
Access to an object in an OODB is performed by manipulating the object using predefined functions, using an explicit query, or by coding explicit references in a programming language.
In the Iris OODB, application developers call functions to retrieve or change values in the object. A program cannot receive a reference to an object which could be passed to other functions. In the GemStone, Vbase, and Orion OODBs, individual objects can be accessed and passed to functions to retrieve or assign values.
In the POSTGRES database, application developers perform queries to retrieve or change values in the object (actually, relational tuples). POSTGRES allows a foreign function to access an object, but as stated above, it must be mapped from the relational data model to the data model of the foreign function's programming language.
Although most OODBs allow the application developer to explicitly retrieve an object from the database (Iris and POSTGRES do not), they do not allow the application developer to specify when objects related to the original object should be retrieved. For example, application developers can access objects in Ontos using one of two modes. In the first mode, an object is explicitly retrieved and referenced objects are implicitly retrieved using an object fault capability. In the other mode, one or more related objects can be explicitly retrieved, but the application must continually check to see if a referenced object is already in memory, and then explicitly retrieve it if it is not. This requires the application developer to employ two completely different models of accessing persistent objects in the same program, which can easily cause errors in the program by the inadvertent and natural use of one mode where the other mode should have been used.
The approach taken by Object Store is quite different from the above OODBs with regard to object access. Object Store's model is more like a persistent memory (an extension of virtual memory computer operating system) than an OODB. Object Design chose to completely reimplement the virtual memory management functions of the C++ programming language and the UNIX (TM) operating system. Whenever a persistent object is created or retrieved from the database, it is installed in a portion of primary memory controlled by Object Design. Thus, references to the object are, in essence, monitored by Object Design's software. If the object is not currently in primary memory, it will be retrieved from the database and installed in primary memory. This style of memory management requires that any class or class library requiring persistence must be written using this memory management scheme, or perform no dynamic memory management thereby resulting in one version of the library for persistent usage and one version for transient usage. Although this approach improves the object storage and retrieval performance, it is inherently dependent on the underlying computer operating system and memory architecture, and thus not portable to other computer systems.
Therefore, these approaches either limit how an application program can access an object, or require additional work in order for the program to access an object.
Most OODBs (except for Iris and DOME) have developed their persistent object storage facility utilizing an existing file management system. They had to develop new implementations of the disk storage structures and management, concurrency control, transaction management, communication, and storage management subsystems. This approach increases the complexity of the overall database system software.
The Iris and DOME OODBs, on the other hand, use existing commercial Relational Database Management Systems (RDBMS) to store their objects. Although the Iris OODB uses Hewlett Packard's relational database HP-SQL, it does not use the SQL interface to that database, restricting access to the objects to the available Iris functions, Iris interactive browser, C language interface, and embedded Iris SQL. Although Iris allows the application developer to define how objects are to be stored, the use of Hewlett Packard's RDBMS imposes a limit on the size of an object. The DOME OODB, which uses Oracle Corporation's Oracle RDBMS, and the POSTGRES system, which has its own relational storage system, decomposes objects into one or more entries in one or more relational tables. This approach requires a relational join whenever more than one attribute value from an object is retrieved. Relational join operations are computationally expensive.
In the GemStone and Object Store OODBs, the unit of concurrency control is not an object but a secondary memory segment, or page. This approach can improve the performance of secondary memory reads and writes, but results in having the storage facility read, write, and lock more data than may be necessary. In addition, this restricts the amount of concurrent access to objects since the OODB system, and not the application developer, chooses the unit of concurrency control.
Most of the OODBs allow related objects to be clustered together in the persistent object storage. GemStone and Orion only allow clustering controls to be specified when the entire database is defined. Vbase and Ontos allow runtime specification of clustering controls to store one persistent object as close as possible to another persistent object. Object Store also allows runtime specification of clustering controls to store statically allocated objects in a specific database and dynamically allocated objects in a specific database or as dose as possible to another persistent object. This requires the application developer to treat similar objects with different models of clustering, which can cause errors in the program by the inadvertent use of one mode where the other mode should have been used. These systems indicate that such clustering specifications are purely hints which the system may ignore. These clustering hints may require rebuilding of the database if they are changed, thereby restricting the ability of the application developers to tune the database's performance by altering the physical grouping of objects. Furthermore, the systems based on relational storage, such as Iris, POSTGRES, and DOME, do not allow user-defined clustering of objects.